What is Single Origin?

Single Origin gives AI agents the context they need to work with your data warehouse.

General-purpose AI models don't know your warehouse: how your tables relate, which columns matter, which jobs build what, or where your compute cost goes. Single Origin builds that knowledge from your warehouse's query history and metadata, as a context graph, and gives it to your agents.

The problem: agents lack data context

An agent asked to fix a slow job or review a SQL change needs evidence: what the job ran last week, what it cost, which tables it reads, who else reads them. Without that, it guesses.

Building this context yourself is hard:

Raw logs are too big and too noisy. Sending months of query history through an LLM is slow and expensive, and most of it is repeats of the same queries.
The insights are buried. Waste hides in patterns across many queries and days: a job that rebuilds its full history every night, a filter that stopped pruning, two tables built the same way. You only see them when the history is organised.

The solution: a context graph for your warehouse

Single Origin groups your queries by job and by shape, links them to the tables they read and write, and tracks their cost and changes day by day. That graph is the context. On top of it, Single Origin:

finds waste and proposes fixes, as recommendations with an estimated daily saving;
measures whether a fix saved money once it ships;
answers your agent's questions with evidence: a query's profile, a job's history, a table's producers and readers.

How it works

flowchart LR
  A[Warehouse query history and metadata] --> B[Context graph]
  B --> C[MCP server]
  B --> D[Web app]
  B --> E[PR review]
  C --> F[Your agent]

Connect. Give Single Origin read-only access to your warehouse's metadata and query history. It never reads the rows in your tables. See Connect your warehouse.
Compile. Single Origin processes your history one day at a time and builds the context graph: jobs, queries, tables, costs and changes.
Expose. Connect your AI agent, such as Claude Code or Cursor, to the Single Origin MCP server. See MCP server.
Execute. Your agent calls Single Origin's tools as it works, so it can explain a slow job, check a fix or review a change with real evidence.